Read the Room: Adapting a Robot's Voice to Ambient and Social Contexts